Research: compile Q8 dirty couplings with a sparse array invoice

[Arham-Tahir64]

The Q8 mixed-target handoff leaves the cost of an actual address action unspecified. This package supplies an in-place eight-block array implementation of every Q8 label, including preparation and restoration of any binary rank-one idempotent register coupling.

The explicit two-star compiler uses exactly 2(wt(u)+wt(v)-2) <= 4(r-1) whole-array XORs, plus four block swaps for -1 or six for a noncentral label. The README proves the arbitrary-dirty-input identity and states the exact emitted bit-XOR and descriptor/storage bill. A regular three-bit affine encoding also calibrates the abstract prices: noncentral Q8 labels require two rank-one affine factors, while -1 is one translation. That rank conclusion is credited to Gareth Davidson/PR1333; this contribution adds the executable bridge and sparse coupling invoice, not a new affine-rank theorem or Q8 word census.

Task: drawn bit-new-register-model, priority revision 6, seed f48632a78b4c1d76bcd9182dda687d24. Base: current canonical main 997e6a9ca7b4f95d50e366f61412fd640d72a88a. TASK.json records the changed scope and bounded pivot from an unsuccessful small cut-invariant scout. SOURCES.json binds the handoff, Gaussian coupling predecessor and affine theorem to exact Git blobs/SHA256; frozen inputs are unchanged. Canonical claim/result.

Validation on native Windows:

• CPython 3.12.10: python scripts/research_package.py --package research/q8-affine-dirty-compiler passes both registered commands, outcome=partial.
• Isolated optimized CPython 3.14.6 verifier and three unittest groups also pass. The receipt covers every idempotent through six roles, all eight labels: 21,336 operators and 969,792 symbolic dirty-input columns, plus 3,840 packet columns.
• Four semantic compiler mutants, three invalid-coupling controls and an affine-underpricing control are enforced with -O. Tests additionally inject actual compiler mutations, reject aliased role containers, and exercise read-only Git-free CLI export, changed/missing receipt and source corruption.
• TMP/TEMP used a workspace-local directory because the sandbox default denied child directory creation. No interpreter or source compatibility shim was needed.

This is a partial local interface result. It supplies no selected-supplier layout, legal chronological caller, routing/hosting/stock or all-size tape/compiler admission, and no new multiplication exponent. The inherited selected record is unchanged. The complete supplier replay, hosted CI, Lean and independent human review are unclaimed. The next bounded task is one named caller and complete bill, compared with its direct alternative; no expanded word search is proposed.

Credit ywz65's mixed-Q8 handoff and dirty-coupling construction; Gareth Davidson/bitplane (PR1333); Aaron Wang (PR1822/2318); Cinna Davis and the retained Q8/register-model lineage. Prepared by Arham Tahir with substantial OpenAI Codex assistance. Apache-2.0; source attribution retained.

Publication commit: 2e69ca272de81e92e91aabac3b4652382b5794f6. check_research_commit.py passes for all 10 package/input files. Final complete overlap snapshot: 261 open PRs (07:16 UTC), 1,892 Discussion records (07:18 UTC); no new same-scope compiler artifact found. Canonical main remained 997e6a9ca7b4f95d50e366f61412fd640d72a88a.